JP5573434B2 - Communication program for information processing apparatus, information processing apparatus, information processing system, and communication method for information processing system - Google Patents

Description

  The present invention relates to a communication program, an information processing apparatus, an information communication system, and a communication method.

  2. Description of the Related Art Conventionally, a system in which a plurality of information processing apparatuses transmit / receive information to / from each other using a network is known. As an example of such a system, there is known a network distributed parallel computer in which a plurality of information processing apparatuses transmit and receive information to and from each other using an MPI (Message Passing Interface) library.

  The information processing apparatus of such a system has a plurality of network interfaces such as HCA (Host Channel Adapter) and NIC (Network Interface Card). The information processing apparatus transmits and receives information to and from other information processing apparatuses using the network interface.

  Under such a configuration, a technique for selecting a network interface to be used for communication from a plurality of network interfaces associated with each other according to the communication status of the own device is known. Hereinafter, processing for selecting a network interface to be used for communication will be described with reference to FIG. FIG. 14 is a diagram for explaining a conventional transmission / reception process. In the example illustrated in FIG. 14, it is assumed that NIC # 1 of information processing apparatus # 1 is associated with NIC # 5 of information processing apparatus # 3.

  First, when transmitting information to the information processing apparatus # 3, the information processing apparatus # 1 selects the network interface of the transmission source from the NICs # 1 to # 2 that the own apparatus has according to the communication status of the own apparatus To do. In the example illustrated in FIG. 14, the information processing apparatus # 1 selects NIC # 1 as the network interface of the transmission source. Then, the information processing apparatus # 1 transmits information to the information processing apparatus # 3 using the NIC # 5 associated with the NIC # 1 as the transmission destination network interface.

  As a technique for transmitting and receiving information to and from other information processing apparatuses using a plurality of network interfaces, a technique called trunking that transmits divided information through independent paths is known. For example, in the example illustrated in FIG. 15, the information processing apparatus # 1 to which the trunking technique is applied divides the data to be transmitted into four data # 1 to # 4 and uses different network interfaces for information Simultaneously transmit to processor # 2. FIG. 15 is a diagram for explaining trunking.

  An information processing apparatus to which the trunking technique is applied is connected to a fat tree type network in which paths (hereinafter referred to as links) connecting network interfaces are independent from each other. For example, in the example illustrated in FIG. 16, each information processing apparatus # 1 to # 2 has a fat tree in which a link connecting NIC # 1 to NIC # 3 and a link connecting NIC # 2 to NIC # 4 are independent. Connected to type network. FIG. 16 is a diagram for explaining a network having a fat tree structure.

JP 2006-121414 A

  However, in the above-described technology for selecting a network interface, a transmission destination network interface is selected according to the communication status of the transmission source, which may cause contention at the transmission destination. That is, since the information processing apparatus selects a network interface to be used for communication regardless of the communication status of the transmission destination, there is a case where a conflict occurs in the network interface of the transmission destination. For this reason, the information processing apparatus has a problem of deteriorating the throughput of the entire system.

  For example, in the example illustrated in FIG. 17, NIC # 3 is associated with NIC # 5, and NC # 5 is in communication. FIG. 17 is a diagram for explaining competition. In the example illustrated in FIG. 17, the information processing apparatus # 2 selects NIC # 3 as the NIC used for communication according to the communication status of the own apparatus.

  Then, the information processing apparatus # 2 selects NIC # 5 associated with NIC # 3 as the destination NIC. For this reason, as indicated by a star in FIG. 17, information processing apparatus # 1 and information processing apparatus # 2 communicate using NIC # 5 simultaneously, and information transmitted to information processing apparatus # 3 is transmitted. Since it concentrates on NIC # 5, contention occurs in NIC # 5.

  As a result, a queue for receiving information is generated in NIC # 5, and NIC # 5 becomes a bottleneck, resulting in a problem that the throughput of the entire system is deteriorated. Note that the information processing apparatus transmits information using a pair of network interfaces associated in advance, so that the combination pattern of network interfaces is limited. As a result, contention due to the use of a specific network interface tends to occur frequently.

  In the trunking technique described above, different information is simultaneously transmitted for each link connecting network interfaces. For this reason, the information processing apparatus has a problem that throughput is deteriorated when each network interface is connected to a network having a structure sharing a link.

  For example, as illustrated in FIG. 18, when each of the information processing apparatuses # 1 to # 2 is connected to the mesh network, a turn waiting for using the shared link indicated by A in FIG. 18 occurs. However, there is a problem that the throughput of the entire system deteriorates. FIG. 18 is a diagram for explaining a mesh network.

  In one aspect, an object of the present invention is to suppress network interface contention.

  According to one aspect, the technology disclosed in the present application is a communication program that selects an interface to be used for communication according to a communication state of each interface. Further, the communication program transmits connection request information to another information processing apparatus using the selected interface. After that, when the communication program receives response information for the connection request information from another information processing apparatus, the communication program specifies the interface used by the other information processing apparatus to transmit the response information from the received response information. To do. Then, the communication program transmits / receives information to / from other information processing apparatuses using the interface used for transmitting the connection request information and the specified interface.

  According to one aspect, suppression of network interface contention can be suppressed.

FIG. 1 is a block diagram for explaining the information processing apparatus according to the first embodiment. FIG. 2 is a diagram for explaining the information processing system according to the second embodiment. FIG. 3 is a block diagram for explaining the computer according to the second embodiment. FIG. 4 is a diagram for explaining the suspended processing. FIG. 5 is a diagram for explaining execution of the suspended processing. FIG. 6 is a diagram for explaining the effect of the computer according to the second embodiment. FIG. 7 is a diagram for explaining connection request information and response information to the connection request information. FIG. 8 is a diagram for explaining transmission and reception of information. FIG. 9 is a diagram for explaining a process of transmitting a Rendezvous request. FIG. 10 is a diagram for explaining processing in which a receiving computer selects a NIC. FIG. 11 is a diagram for explaining processing in which the computer according to the second embodiment transmits and receives information. FIG. 12 is a sequence diagram for explaining the flow of processing executed by the computer according to the second embodiment. FIG. 13 is a diagram for describing an example of a computer that executes a communication program. FIG. 14 is a diagram for explaining a conventional transmission / reception process. FIG. 15 is a diagram for explaining trunking. FIG. 16 is a diagram for explaining a network having a fat tree structure. FIG. 17 is a diagram for explaining competition. FIG. 18 is a diagram for explaining a mesh network.

  A communication program, an information processing apparatus, an information communication system, and a communication method according to the present application will be described below with reference to the accompanying drawings.

  In the following first embodiment, an example of an information processing apparatus will be described with reference to FIG. FIG. 1 is a block diagram for explaining the information processing apparatus according to the first embodiment. The information processing apparatus is an information processing apparatus that communicates with at least an information processing apparatus that executes a similar communication program.

  As illustrated in FIG. 1, the information processing apparatus 1 includes a first selection unit 2, a connection request transmission unit 3, a specification unit 4, a communication unit 5, a second selection unit 6, a response transmission unit 7, and interfaces 8a to 8c. . The information processing apparatus 1 communicates with the information processing apparatus 1a having the same function as the information processing apparatus 1 via a network. In addition, each part 2a-7a, 8d-8f which the information processing apparatus 1a has shall exhibit the function similar to each part 2-7, 8a-8c which the information processing apparatus 1 has, and about each part of the information processing apparatus 1a Description is omitted.

  The first selection unit 2 selects an interface to be used for communication according to the communication status of each interface 8a to 8c. The connection request transmission unit 3 transmits the connection request information to the information processing apparatus 1a using the interface selected by the first selection unit 2.

  When the identification unit 4 receives response information for the connection request information transmitted by the connection request transmission unit 3 from the information processing apparatus 1a, the information processing apparatus 1a transmits the response information from the received response information. Identify the interface used. The communication unit 5 transmits and receives information to and from the information processing apparatus 1a using the interface used by the connection request transmission unit 3 to transmit connection request information and the interface specified by the specifying unit 4.

  That is, when the information processing apparatus 1 is the transmission side and the information processing apparatus 1a is the reception side, the information processing apparatus 1 that is the transmission side transmits connection request information using an interface convenient for the own apparatus. To do. Then, the information processing apparatus 1 identifies the interface selected by the information processing apparatus 1a on the reception side, and communicates using an interface convenient for the identified reception side.

  When receiving the connection request information from the information processing apparatus 1a, the second selection unit 6 selects an interface to be used for communication with the information processing apparatus 1a according to the communication status of each interface 8a to 8c. The response transmission unit 7 transmits the response information for the connection request information to the information processing apparatus 1a using the interface selected by the second selection unit 6.

  That is, when the information processing apparatus 1 is the receiving side and the information processing apparatus 1a is the transmitting side, the information processing apparatus 1 that is the receiving side selects an interface that is convenient for the receiving side and uses the selected interface. Then, the response information is notified to the information processing apparatus 1a.

  As described above, the information processing apparatus 1 transmits the connection request information to the information processing apparatus 1a using the interface selected according to the communication state of the own apparatus. In addition, when the response information for the transmitted connection request information is received from the information processing device 1a, the information processing device 1 identifies the interface used by the information processing device 1a from the received response information. Then, the information processing apparatus 1 transmits / receives information to / from the information processing apparatus 1a using the interface used for transmitting the connection request information and the identified interface.

  Further, when receiving the connection request information, the information processing apparatus 1a selects an interface used for communication with the information processing apparatus 1 according to the communication state of each of the interfaces 8d to 8f. Then, the information processing apparatus 1a transmits response information using the selected interface.

  For this reason, the information processing apparatus 1 can select the interface used for communication according to the communication state of the information processing apparatus 1a that is the information transmission destination. Further, the information processing apparatus 1 can transmit and receive information without limiting the interface used for communication. For this reason, the information processing apparatus 1 determines the network interface to be used in accordance with the communication status between the transmission side and the reception side, thereby suppressing network interface contention. Moreover, the information processing apparatus 1 improves the throughput.

  In the second embodiment, an information processing system having a plurality of computers that execute communication programs will be described.

  First, an information processing system according to the second embodiment will be described with reference to FIG. FIG. 2 is a diagram for explaining the information processing system according to the second embodiment. As illustrated in FIG. 2, the information processing system includes a plurality of computers 20 to 25. Each of the computers 20 to 25 has four NICs (Network Interface Cards). The NICs 30 to 50 connect the computers 20 to 25 via a network. The computers 20 to 25 are connected to the NICs 30 to 50 via a network.

  Each computer 20-25 shall perform the same process by running the same program. In the following description, each part of the computer 20 will be described, and description of the other computers 21 to 25 will be omitted.

  Next, each part of the computer 20 will be described with reference to FIG. FIG. 3 is a block diagram for explaining the computer according to the second embodiment. In the example illustrated in FIG. 3, the computer 20 includes four NICs 30 to 33. The computer 20 includes a process execution unit 10 and an information storage unit 16. Each part will be described below.

  The information storage unit 16 temporarily stores information to be transmitted to the other computers 21 to 25. The information storage unit 16 temporarily stores information received from the other computers 21 to 25 using the NICs 30 to 33.

  The NIC 30 is connected to each of the NICs 34 to 53 included in the other computers 21 to 25 via a network, and controls communication processing with each of the NICs 34 to 53. Specifically, when the NIC 30 receives connection request information from each of the NICs 34 to 53 of the other computers 21 to 25, the NIC 30 transfers the received connection request information to the response transmission unit 13.

  For example, when receiving connection request information from the NIC 50 of the computer 25, the NIC 30 transfers the received connection request information to the response transmission unit 13. In addition, when receiving information from the NIC 51 of the computer 25, the NIC 30 transfers the received information to the communication unit 14.

  In addition, when the NIC 30 acquires function call information for transmitting response information to the NIC 51 of the computer 25 from the holding unit 15, the NIC 30 transmits the response information with the NIC 51 of the computer 25 as a transmission destination. Further, when the NIC 30 acquires function call information for transmitting information from the holding unit 15 using the NIC 51 of the computer 25, the NIC 30 transmits information to the NIC 51 of the computer 25.

  The process execution unit 10 executes a communication program according to the second embodiment. Specifically, the process execution unit 10 includes a selection unit 11, a connection request transmission unit 12, a response transmission unit 13, a communication unit 14, and a holding unit 15. And the process execution part 10 performs a communication program using each part 11-15, and performs transmission / reception of the information with the other computers 21-25.

  In the following description, functions of each unit included in the process execution unit 10 will be described. Further, in the following description, an example will be described in which the process execution unit 10 transmits information held by the computer 20 to another computer 25 using an MPI (Message Passing Interface) library.

  The holding unit 15 temporarily holds the communication reservation for each NIC. In addition, the holding unit 15 executes processing indicated by the communication reservation that is initially reserved among the communication reservations related to the NIC that has finished communication. Here, the communication reservation is information for identifying a function call necessary for causing each of the NICs 30 to 33 to execute communication processing, such as “MPI_Isend” or “MPI_Irecv”.

  Specifically, the holding unit 15 receives the communication reservation transmitted from the connection request transmission unit 12, the response transmission unit 13, and the communication unit 14. Then, the holding unit 15 temporarily holds the received communication reservation when the NIC executing the communication reservation is communicating. Thereafter, when the NIC that executes the received communication reservation ends communication, the holding unit 15 is a communication reservation that uses the NIC that has ended communication among the reserved communication reservations, and the first The communication reservation stored in is notified to the NIC. Then, when the communication process related to the notified communication reservation is completed, the holding unit 15 deletes the communication reservation related to the completed communication process.

  For example, in the example illustrated in FIG. 4, the holding unit 15 stores one communication reservation for the NIC 30, two communication reservations for the NIC 31, three communication reservations for the NIC 32, and four communication reservations for the NIC 33. In addition, as illustrated by the dotted line in FIG. 4, when the communication reservation executed by the NIC 30 is newly received, the holding unit 15 newly stores the communication reservation in association with the NIC 30. FIG. 4 is a diagram for explaining the suspended processing.

  In the example shown in FIG. 5, the holding unit 15 has three communication reservations using the NIC 30, one communication reservation using the NIC 31, two communication reservations using the NIC 32, and a communication reservation using the NIC 33. Is stored. FIG. 5 is a diagram for explaining execution of the suspended processing. In this example, when the communication process being executed by the NIC 30 is completed, the holding unit 15 deletes the communication reservation executed by the NIC 30 as indicated by the dotted line in FIG.

  Here, when executing the communication reservation of the NIC 30, the holding unit 15 transmits information identifying the function call included in the communication reservation to the NIC 30. When the NIC 30 acquires information for identifying a function call from the holding unit 15, the NIC 30 executes a process according to the information for identifying the acquired function call.

  As described above, when the NIC 30 to 33 receive the communication reservation generated by the connection request transmission unit 12, the response transmission unit 13, and the communication unit 14 while the NICs 30 to 33 are communicating, the reservation unit 15 temporarily holds the communication reservation. To do. When the NIC used by the reserved communication reservation can finish the communication process and execute the communication process related to the new communication reservation, the hold unit 15 performs the communication process related to the reserved communication reservation. Is executed. The holding unit 15 stores the held processing as information for identifying a function call, and uses the address of the memory in which the communication reservation is stored for communication reservation order management.

  Returning to FIG. 3, the selection unit 11 selects a NIC to be used for communication according to the communication status of each of the NICs 30 to 33. In addition, when the connection request information is received from the other computers 21 to 25, the selection unit 11 selects a NIC to be used for communication with the other computers 21 to 25 according to the communication status of each NIC 30 to 33. To do.

  Moreover, the selection part 11 selects NIC used for transmission of the information to the other computers 21-25 according to the communication reservation for using each NIC30-33. In addition, the selection unit 11 selects a NIC used for receiving information from the other computers 21 to 25 in accordance with a communication reservation for using each of the NICs 30 to 33.

  Specifically, when receiving an instruction to select a NIC from the connection request transmission unit 12 to be described later, the selection unit 11 selects the NIC with the smallest communication reservation held by the holding unit 15. Then, the selection unit 11 notifies the connection request transmission unit 12 of the selected NIC.

  Similarly, when the selection unit 11 receives an instruction to select a NIC from the response transmission unit 13 described later, the selection unit 11 selects the NIC with the smallest communication reservation reserved by the reservation unit 15. Then, the selection unit 11 notifies the response transmission unit 13 of the selected NIC.

  Here, the process performed by the selection unit 11 will be described using the example of FIG. For example, when the selection unit 11 acquires an instruction to select a NIC from the connection request transmission unit 12 to be described later, the selection unit 11 identifies a process of using each NIC 30 to 33 held by the holding unit 15.

  Here, in the example illustrated in FIG. 4, the holding unit 15 stores one communication reservation for the NIC 30, two communication reservations for the NIC 31, three communication reservations for the NIC 32, and four communication reservations for the NIC 33. . For this reason, the selection unit 11 selects the NIC 30 with the smallest suspended processing. That is, the selection unit 11 selects the NIC 30 as the NIC used for communication. Then, the selection unit 11 notifies the NIC 30 to the connection request transmission unit 12.

  In addition, when the selection unit 11 acquires an instruction to select a NIC from the response transmission unit 13, the selection unit 11 selects the NIC 30 as a NIC to be used for communication based on the number of communication reservations held by the holding unit 15. To do. Thereafter, the selection unit 11 notifies the NIC 30 to the connection request transmission unit 12.

  Returning to FIG. 3, the connection request transmission unit 12 transmits the connection request information to another computer using the NIC selected by the selection unit 11. Specifically, the connection request transmission unit 12 transmits an instruction to select the NIC to the selection unit 11 when a notification to transmit information is acquired from the communication unit 14. Further, as will be described later, the connection request transmission unit 12 is notified from the communication unit 14 of a computer that is an information transmission destination.

  When the connection request transmission unit 12 is notified of the NIC used for communication from the selection unit 11, the NIC associated with the NIC notified from the selection unit 11 among the NICs of the transmission destination computer. Is determined. In addition, the connection request transmission unit 12 generates connection request information including the NIC address notified from the selection unit 11. Then, the connection request transmission unit 12 transmits a process of transmitting connection request information using the NIC notified from the selection unit 11 and the determined NIC to the reservation unit 15 as a communication reservation.

  For example, the connection request transmission unit 12 transmits an instruction to select the NIC to the selection unit 11 when a notification that the information is transmitted to the computer 25 is acquired from the communication unit 14. Further, the connection request transmission unit 12 is notified of the NIC 30 as the NIC used for communication from the selection unit 11.

  When the NIC 30 is notified as the NIC used for communication, the connection request transmission unit 12 determines the NIC 50 that is the NIC associated with the notified NIC 30 among the NICs 50 to 53 included in the computer 25. Further, the connection request transmission unit 12 generates connection request information including the address of the NIC 30 as connection request information. Then, the connection request transmission unit 12 generates a communication reservation indicating processing for transmitting the generated connection request information to the computer 25 using the NIC 30 and the NIC 50. Thereafter, the connection request transmission unit 12 transmits the generated communication reservation to the holding unit 15.

  The response transmission unit 13 transmits the response information for the connection request information to the transmission source of the connection request information using the NIC selected by the selection unit 11. In addition, the response transmission unit 13 transmits response information including an address that uniquely identifies the NIC selected by the selection unit 11.

  Specifically, when the response transmission unit 13 receives the connection request information transmitted from the other computers 20 to 25, the response transmission unit 13 transmits an instruction to select the NIC to the selection unit 11. Further, the response transmission unit 13 determines the NIC of the transmission source used for transmitting the received connection request information based on the NIC address included in the received connection request information.

  Further, when the selection unit 11 is notified of the NIC used for communication, the response transmission unit 13 generates response information including the NIC address notified from the selection unit 11. And the response transmission part 13 produces | generates the communication reservation which shows the process which transmits the produced | generated response information using NIC notified from the selection part 11, and the NIC which the transmission origin of the connection request used.

  Thereafter, the response transmission unit 13 transmits the generated communication reservation to the holding unit 15. In addition, when the response transmission unit 13 receives the connection request information, the response transmission unit 13 transmits a notification that information is transmitted from another computer to the communication unit 14.

  Here, an example of processing executed by the response transmission unit 13 will be described. In the following example, the response transmission unit 13 receives connection request information transmitted from the computer 25. Further, it is assumed that the computer 25 transmits connection request information using the NIC 51 and the NIC 31. For example, when the connection request information transmitted from the computer 25 is received via the NIC 31, the response transmission unit 13 transmits an instruction to select the NIC to the selection unit 11.

  Moreover, the response transmission part 13 discriminate | determines NIC51 of the computer 25 used in order to transmit the received connection request information based on the received response information. Further, when the NIC 30 is notified from the selection unit 11 as the NIC used for communication, the response transmission unit 13 generates response information including the address of the NIC 30. And the response transmission part 13 produces | generates the communication reservation which shows the process which transmits the produced | generated response information to the computer 25 using NIC30 and NIC51. Thereafter, the response transmission unit 13 transmits the generated communication reservation to the holding unit 15.

  When the communication unit 14 receives response information for the connection request information transmitted by the connection request transmission unit 12 from the other computers 21 to 25, the other computers 21 to 25 respond from the received response information. Identify the NIC that was used to send the information. Further, when the response unit 14 receives the response information from the other computers 21 to 25, the communication unit 14 uniquely identifies the NIC selected by the selection unit related to the other computers 21 to 25 from the received response information. Identify the address.

  And the communication part 14 specifies NIC used in order that the computer which transmitted response information may transmit response information from the identified address. And the communication part 14 transmits / receives information with the other computers 21-25 using the NIC used for the connection request transmission part 12 transmitting connection request information, and identified NIC.

  Specifically, when the communication unit 14 transmits the information stored in the information storage unit 16 to any of the other computers 21 to 25, the communication unit 14 notifies the connection request transmission unit 12 that the information is transmitted. Send. In addition, the communication unit 14 notifies the connection request transmission unit 12 of the computer that is the transmission destination of information.

  Further, the communication unit 14 receives response information for the connection request information transmitted from the computer 20 from another computer. Further, when receiving the response information, the communication unit 14 based on the address of the NIC included in the received response information, the source NIC used to transmit the received response information Is determined.

  Then, the communication unit 14 performs processing for transmitting the information stored in the information storage unit 16 to the destination computer using the NIC of the computer 20 that has received the response information and the determined NIC of the transmission source. To do. That is, the communication unit 14 generates a communication reservation indicating processing, and transmits the generated communication reservation to the holding unit 15.

  Further, when the communication unit 14 obtains a notification that information is transmitted from another computer from the response transmission unit 13, the communication unit 14 secures an area for storing the transmitted information in the information storage unit 16. To do. Then, when any of the NICs 30 to 33 receives information transmitted from another computer, the communication unit 14 acquires the received information and is acquired in an area secured in the information storage unit 16. Stored information.

  Here, although the example of the process which the communication part 14 performs is demonstrated, the communication part 14 shall receive response information from the computer 25 in the following examples. Further, it is assumed that the computer 25 transmits the response information using the NIC 51 and the NIC 30.

  For example, when the communication unit 14 receives response information for the connection request information transmitted from the computer 20 to the computer 25 using the NIC 30, the communication unit 14 determines the transmission source information based on the address of the NIC included in the response information. The NIC 51 is determined. And the communication part 14 produces | generates the communication reservation which shows the process which transmits the information memorize | stored in the information storage part 16 to the computer 25 using NIC30 and NIC51. Thereafter, the communication unit 14 transmits the generated communication reservation to the holding unit 15.

  Further, when the communication unit 14 obtains a notification that information is transmitted from the computer 25 from the response transmission unit 13, the communication unit 14 secures an area for storing the transmitted information in the information storage unit 16. . When the NIC 31 receives information transmitted from the computer 25, the communication unit 14 acquires the received information and stores the acquired information in an area secured in the information storage unit 16.

  As described above, when transmitting information to the computer 25, the computer 20 transmits connection request information to the computer 25 using the NIC that is most suitable for information transmission / reception processing among the NICs 30 to 33 of its own device. . Further, the computer 25 transmits the response information corresponding to the connection request information to the computer 20 using the NIC most suitable for information transmission / reception processing among the NICs 50 to 53 of the own device. When the computer 20 receives the response information from the computer 25, the computer 20 uses the NIC used by the own device for transmitting the connection request information and the NIC used by the computer 25 for transmitting the response information. The information transmission / reception process is executed.

  That is, the computer 20 selects the NIC used for communication in consideration of not only the communication status of the own device but also the communication status of the computer 25. For this reason, the computer 20 can execute the information transmission / reception process using the NIC of the own apparatus optimal for the information transmission / reception process and the destination NIC optimal for the information transmission / reception process. As a result, the computer 20 can reduce the possibility of contention in the destination NIC, thereby improving the overall system throughput.

  Further, the computer 20 can execute transmission / reception of information without limiting the combination of NICs used for transmitting / receiving information. For example, when two computers having N NICs transmit and receive information, the number of NIC combinations used is a power of N. For this reason, as shown in FIG. 6, the computer 20 has many combinations of interfaces, and can suppress competition of network interfaces. As a result, the computer 20 can improve the throughput of the entire system. FIG. 6 is a diagram for explaining the effect of the computer according to the second embodiment.

  Next, a specific example of transmitting / receiving connection request information and response information to the connection request information will be described with reference to FIG. FIG. 7 is a diagram for explaining connection request information and response information to the connection request information. In the example illustrated in FIG. 7, the computer 20 and the computer 25 execute an application program that executes message communication using the Rendezvous protocol.

  This application program is executed by the process execution unit 10 described above. In the example illustrated in FIG. 7, the computer 20 and the computer 25 transmit and receive connection request information and response information using the MPI library. Of the NICs included in each computer, the NICs 32 to 33 and 52 to 53 are not shown.

  First, the application program executed on the computer 20 transmits a Rendezvous request from the NIC 30 to the computer 25 by calling “MPI_Isend” as connection request information. When the application program executed by the computer 25 receives the Rendezvous request transmitted from the NIC 30, it creates a data area on the memory.

  Further, the application program executed on the computer 25 selects the NIC 51 that is an appropriate NIC for transmitting and receiving information to the own apparatus. Then, the application program executed on the computer 25 calls “MPI_Irecv” and transmits a Rendezvous response to the computer 20 as response information.

  At this time, the computer 25 sends a Rendezvous response from the NIC 51 to notify the application program of the computer 20 of an appropriate NIC for the computer 25 to receive information.

  As described above, when receiving the Rendezvous request from the transmission-side application program, the reception-side application executes preparation for receiving information, that is, creation of the data area and selection of the NIC to be used. . Then, when preparation for receiving information is completed, the receiving-side application transmits a Rendezvous response to the Rendezvous request transmission source.

  Next, the flow of processing for transmitting information of the computer 20 to the computer 25 will be described with reference to FIG. FIG. 8 is a diagram for explaining transmission and reception of information. First, the computer 20 selects its own NIC 31 as the NIC for transmitting the Rendezvous request. Next, the computer 20 transmits a Rendezvous request from the NIC 31. Here, in the Rendezvous request, “Send NIC: 31” which is the address of the NIC 31 and “Send Address: X” which is information indicating the memory address where the transmitted information is stored are stored.

  When the computer 25 receives the Rendezvous request transmitted from the NIC 31, the computer 25 selects the NIC 52 of its own device as the NIC for receiving information. Then, the computer 25 transmits “GET” from the NIC 52 as a Rendezvous response.

  Here, “GET” transmitted as the Rendezvous response stores “Recv NIC: 52” which is the address of the NIC 52 selected by the computer 25 in addition to the information stored in the Rendezvous request. The “GET” transmitted as the Rendezvous response stores the memory address “Recv Address: Y” of the computer 25 that stores the transmitted information.

  When the computer 20 receives “GET” transmitted as a Rendezvous response, the computer 20 transmits information on the own device to the computer 25 in accordance with the information stored in the received “GET”. That is, the computer 20 uses the NIC 31 and the NIC 52 to transmit the data stored in the memory address X of the own device to the memory address Y of the computer 25.

  Next, the flow of communication processing executed by the computer according to the second embodiment will be described with reference to FIGS. First, processing for transmitting a Rendezvous request will be described with reference to FIG. FIG. 9 is a diagram for explaining a process of transmitting a Rendezvous request.

  In the example illustrated in FIG. 9, when the transmission function “MPI_Isend” is called in the process, the computer 20 selects the NIC 30 that is vacant among the NICs 30 to 33 included in the computer 20. Then, the computer 20 uses the NIC 30 to transmit a Rendezvous request including the memory address information of the information transmitted by “MPI_Isend” and the address of the NIC 30.

  Next, a process of selecting a NIC by the receiving computer will be described with reference to FIG. FIG. 10 is a diagram for explaining processing in which a receiving computer selects a NIC. In the example illustrated in FIG. 10, when “MPI_Irecv” corresponding to “MPI_Isend” called by the computer 20 is called, the computer 25 receives the Rendezvous request transmitted by the computer 20. Next, the computer 25 selects the vacant NIC 51.

  Next, the process in which a computer transmits / receives information is demonstrated using FIG. FIG. 11 is a diagram for explaining processing in which the computer according to the second embodiment transmits and receives information. In the example illustrated in FIG. 11, the computer 25 transmits “GET” to the computer 20 using the NIC 51 and the NIC 30 which is the most suitable NIC for the computer 20 to transmit information.

  Then, the computer 25 receives information using the NIC 30 and the NIC 51 as indicated by a dotted line in FIG. “GET” includes information indicating the NIC 30, an address of information transmitted from the computer 20, information indicating the NIC 51, and an address where the computer 25 stores information. In the example shown in FIG. 11, the computer 25 receives information by RDMA (Remote Direct Memory Access) transfer by “GET”.

[Computer processing flow]
Next, the flow of processing executed by the computer according to the second embodiment will be described with reference to FIG. FIG. 12 is a sequence diagram for explaining the flow of processing executed by the computer according to the second embodiment. In the example illustrated in FIG. 12, the computer 20 starts processing with “MPI_Isend” being called by a process to be executed as a trigger. In addition, the computer 25 starts processing with “MPI_Irecv” being called by a process to be executed as a trigger.

  First, the computer 20 searches for a free NIC suitable for transmitting information (step S101). Next, the computer 20 adds a communication reservation indicating processing for transmitting a Rendezvous request (step S102). Next, the computer 20 executes a process of transmitting a Rendezvous request related to communication reservation (step S103).

  On the other hand, when “MPI_Irecv” is called, the computer 25 waits for reception of a Rendezvous request by the corresponding “MPI_Isend” (step S104). Then, when receiving the Rendezvous request, the computer 25 searches for a free NIC of its own device (step S105).

  Next, the computer 25 adds a communication reservation indicating processing for transmitting “GET” as a Rendezvous response to the communication reservation list (step S106). Then, the computer 25 instructs GET to the selected empty NIC (step S107), and transmits a GET control signal to the computer 20 (step S108).

  When the computer 20 receives the control signal (step S109), the computer 20 sets the I / F (Interface) (step S110). That is, the computer 20 performs setting for transmitting information using the NIC selected in step S101 and the NIC selected in step S105.

  Next, the computer 20 transmits data to the computer 25 using the NIC selected in step S101 and the NIC selected in step S105 (step S111). Further, the computer 25 receives data transmitted using the NIC selected in step S101 and the NIC selected in step S105 (step S112).

  When the computer 25 receives the data, the computer 25 deletes the communication reservation related to the executed Rendezvous response (step S113). Further, when the data is transmitted, the computer 20 deletes the communication reservation related to the executed Rendezvous request (step S114).

[Effect of Example 2]
As described above, when communicating with the computer 25, the computer 20 according to the second embodiment selects a NIC to be used for communication according to the communication status of the NICs 30 to 31. Then, the computer 20 transmits connection request information from the selected NIC to the computer 25. Further, when the computer 25 receives the connection request information, the computer 25 selects a NIC to be used for communication, and transmits a response to the connection request information to the computer 20 using the selected NIC.

  When the computer 20 receives response information for the transmitted connection request information from another computer, the computer 20 identifies the NIC used by the computer 25 from the received response information. Thereafter, the computer 20 transmits information to the computer 25 using the NIC selected by the own device and the identified NIC.

  Therefore, the computer 20 can select a NIC corresponding to the communication status of the computer 25 when transmitting information to the computer 25. That is, a convenient NIC for performing communication can be determined according to the communication status between the transmission side and the reception side. As a result, the computer 20 can suppress network interface contention in the computer 25. Further, the computer 20 can improve the throughput of the entire system. Further, since the computer 20 can use all the NICs included in the computer 20 and the computer 25, it is possible to suppress network interface contention and improve the throughput of the entire system.

  In addition, since the computer 20 can suppress network interface contention regardless of the system structure, the overall throughput of the system can be improved.

  Further, the computer 20 selects an interface to be used for communication with the computer 25 in accordance with a communication reservation for using the interface. Further, the computer 20 selects a NIC used for receiving information from the computer 25 in accordance with a communication reservation for using the NIC. Therefore, the computer 20 can appropriately select the NIC used for communication, and as a result, the throughput of the entire system can be improved.

  In addition, the computer 25 transmits response information including the address of the NIC selected by the own device. For this reason, since the computer 20 can select an appropriate NIC used for communication, contention of the network interface can be suppressed and the throughput of the entire system can be improved.

  In addition, each time a message is communicated, the sending side and the receiving side check each other for an appropriate network interface, and the information is shared with each other because communication for information sharing is required separately, resulting in increased overhead. Will degrade the throughput.

  However, each of the computers 20 to 25 uses the exchange of the Rendezvous request and the Rendezvous response to notify the NIC that the own device wants to use for communication, so that information can be shared with very little overhead. As a result, each of the computers 20 to 25 efficiently equalizes the network interface usage frequency and improves the throughput of the entire system.

  In the description of the present embodiment, communication between the computer 20 and the computer 25 has been described. However, it goes without saying that the invention disclosed in the present embodiment can be applied not only to communication between the computer 20 and the computer 25 but also to communication between arbitrary computers and has the same effect.

  Although the embodiments of the present invention have been described so far, the embodiments may be implemented in various different forms other than the embodiments described above. Therefore, another embodiment included in the present invention will be described below as a third embodiment.

(1) About NIC In the embodiment described above, a computer having a NIC (Network Interface Card) has been described as an example of an information processing apparatus having a network interface. However, the embodiment is not limited to this. For example, the network interface may be an HCA (Host Channel Adapter) or an SCSI (Small Computer System Interface).

(2) Protocol In the embodiment described above, a computer that performs communication using the Rendezvous protocol has been described. However, the embodiment is not limited to this. For example, the present invention can be applied to any protocol that exchanges connection request information and response information between information processing apparatuses that perform communication.

(3) Regarding the holding unit In the embodiment described above, the NIC used for communication is selected according to the number of communication reservations temporarily held by the holding unit 15. However, the embodiment is not limited to this. For example, the time that the processing indicated by each communication reservation will use the NIC may be calculated, and the NIC with the shortest calculated time may be selected. . Further, the amount of information exchanged by the processing indicated by the communication reservation may be calculated, and the NIC with the smallest calculated information amount may be selected.

(4) Program By the way, the information processing apparatus according to the first embodiment and the computer shown in the second embodiment described the case where various processes are realized using hardware. However, the embodiment is not limited to this, and may be realized by executing a program prepared in advance by an information processing apparatus or a computer. In the following, an example of a computer that executes a program having the same function as that of the information processing apparatus described in the first embodiment will be described with reference to FIG. FIG. 13 is a diagram for describing an example of a computer that executes a communication program.

  In the computer 200 illustrated in FIG. 13, a RAM (Random Access Memory) 120, a ROM (Read Only Memory) 130, and an HDD (Hard Disk Drive) 150 are connected by a bus 170. Further, the computer 200 illustrated in FIG. 13 is connected to a CPU (Central Processing Unit) 140 via a bus 170. Further, I / O (Input Output) 160 to 162 which are a plurality of interfaces for connecting to the network are connected to the bus 170.

  In the ROM 130, a first search program 132, a connection request transmission program 133, a specifying program 134, a communication program 135, a second selection program 136, and a response transmission program 137 are stored in advance. When the CPU 140 reads out the programs 132 to 137 from the ROM 130 and executes them, the programs 132 to 133 function as the first search process 142 and the connection request transmission process 143 in the example shown in FIG.

  The programs 134 to 137 function as a specific process 144, a communication process 145, a second selection process 146, and a response transmission process 147. In addition, each process 142-147 exhibits the function similar to each part 2-7 shown in FIG. Moreover, each process 142-147 can also be made to exhibit the function equivalent to each part which concerns on Example 2. FIG.

  The communication program described in this embodiment can be realized by executing a program prepared in advance on a computer such as a personal computer or a workstation. This program can be distributed via a network such as the Internet.

  The program is recorded on a computer-readable recording medium such as a hard disk, a flexible disk (FD), a CD-ROM (Compact Disc Read Only Memory), an MO (Magneto Optical Disc), a DVD (Digital Versatile Disc). The The program can also be executed by being read from a recording medium by a computer.

DESCRIPTION OF SYMBOLS 1 Information processing apparatus 2 1st selection part 3 Connection request transmission part 4 Identification part 5 Communication part 6 2nd selection part 7 Response transmission part 8a-8c Interface 10 Process execution part 11 Selection part 12 Connection request transmission part 13 Response transmission part 14 Communication unit 15 Reservation unit 16 Information storage unit 20 Computer 30 to 33 NIC

Claims (6)

In a communication program for an information processing apparatus having a plurality of interfaces,
In the information processing apparatus,
A first selection procedure for selecting an interface to be used for communication according to the communication status of each interface;
A connection request transmission procedure for transmitting connection request information to another information processing apparatus using the interface selected by the first selection procedure;
When receiving the response information to the connection request information transmitted by the connection request transmission procedure from the other information processing apparatus, from said response information has been received, said other information processing apparatus transmits the response information Specific steps to identify the interface used to
Using the interface, wherein the connection request transmission procedure is used to transmit the connection request information, and an interface that is specified by the specifying step, a communication procedure for transmitting and receiving said information processing apparatus and information,
When receiving the connection request information from the other information processing apparatus, a second selection procedure that selects an interface according to the communication status of each interface is used to communicate with the other information processing apparatus,
A response transmission procedure for transmitting response information for the received connection request information to a transmission source of the received connection request information using the interface selected by the second selection procedure;
The communication program of an information processing apparatus, characterized in that to perform. In the communication program of the information processing apparatus,
The first selection procedure selects an interface used for transmitting information to the other information processing apparatus according to a communication reservation for using each of the interfaces.
The said 2nd selection procedure selects the interface used for reception of the information from said other information processing apparatus according to the communication reservation for using each said interface. Program of information processing apparatus . In the communication program of the information processing apparatus,
The specific procedure, when receiving the response information to the connection request information transmitted by the connection request transmission procedure from the other information processing apparatus, from said response information has been received, the other information processing apparatus from the specific information which the other information processing apparatus that is selected according to the communication status of each interface the interface used to communicate with the information processing apparatus identifies a specific information specific was identify with, the Identify the interface that other information processing devices used to send response information,
The response transmitting step, the communication of the information processing apparatus according to claim 1 or 2, characterized in that transmitting the said response information including the identification information of the interface selected by the second selection procedure for specific program. Multiple interfaces,
A first selection unit that selects an interface to be used for communication according to the communication status of each interface;
A connection request transmission unit that transmits connection request information to another information processing apparatus using the interface selected by the first selection unit;
When receiving the response information to the connection request information transmitted by the connection request transmission unit from the other information processing apparatus, from said response information has been received, it said other information processing apparatus transmits the response information A specific part that identifies the interface used for
An interface for the connection request transmission unit is used for transmitting the connection request, a communication unit using the specified interface by the specifying unit, to transmit and receive the other information processing apparatus and information,
When receiving the connection request information from the other information processing apparatus includes a second selection unit that selects an interface according to the communication status of each interface is used to communicate with the other information processing apparatus,
And characterized in that it has a response transmitting unit using an interface selected by the second selecting unit, and transmits the response information in response to the connection request information received to the sender of the received connection request information Information processing apparatus. A first information processing apparatus and a second information processing apparatus that communicate with each other;
The first information processing apparatus includes:
Multiple interfaces,
A first selection unit that selects an interface to be used for communication with the second information processing apparatus according to the communication status of each interface;
A connection request transmission unit that transmits connection request information to the second information processing apparatus using the interface selected by the first selection unit;
When receiving the response information in response to the connection request the connection request information transmitted by the transmission unit from said second information processing apparatus, from said response information has been received, the second information processing apparatus response information A specific part that identifies the interface used to send,
A communication unit that transmits / receives information to / from the second information processing apparatus using the interface used by the connection request transmitting unit to transmit the connection request information and the interface specified by the specifying unit; ,
Have
The second information processing apparatus
Multiple interfaces,
When connection request information is received from the first information processing apparatus, a second selection unit that selects an interface to be used for communication with the first information processing apparatus according to the communication status of each interface;
Using an interface selected by the second selecting unit, and having a response transmitting unit that transmits the response information in response to the received connection request information to the sender of the received connection request information Information processing system . A communication method of an information processing system having a plurality of interfaces and having a first information processing device and a second information processing device communicating with each other,
A first selection step in which the first information processing apparatus selects an interface to be used for communication with the second information processing apparatus according to a communication status of each interface;
A connection request transmission step of transmitting connection request information to the second information processing apparatus using the interface selected in the first selection step;
When receiving the response information in response to the connection request transmission connection request information transmitted at step from said second information processing apparatus, from said response information has been received, the second information processing apparatus response information Specific steps to identify the interface used to send,
An interface for the connection request transmission step is used to transmit the connection request information, using the specified interface by the specifying step, a communication step of performing transmission and reception of the second information processing apparatus and information Perform a method that includes
When the second information processing apparatus receives connection request information from the first information processing apparatus, an interface used for communication with the first information processing apparatus is selected according to the communication status of each interface. A second selection step to select;
Performing a response transmission step of transmitting response information for the received connection request information to a transmission source of the received connection request information using the interface selected by the second selection step. A communication method of an information processing system characterized by the above.

Images